SAO Parameter Encoding with Dynamic Band Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The current Sample Adaptive Offset (SAO) method in H.265/HEVC has limitations in fine-tuning pixel values, leading to suboptimal image quality and excessive bit transmission for SAO parameters, particularly around median gray-scale values and in next-generation video codecs.
Innovation Solution
The proposed method divides the gray-scale range into classes with a predetermined number of levels, counts pixel distribution, selects the top n classes with the most pixels, and adds offset values to these classes, reducing the number of bits required for SAO parameters by signaling only the added offset values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If the currently utilized band offset with wide bandwidth is applied to specific consecutive gray-scale values, then the pixel values can be adjusted to reduce compression error, but it becomes difficult to perform fine adjustment of pixel values, limiting image quality improvement
Solution Approach 1:
The gray-scale range is divided into multiple bands (e.g., 32 bands), and instead of applying offset to all bands uniformly, the invention selects only 4 consecutive bands that require adjustment. This segmentation allows precise targeting of specific gray-scale regions needing fine adjustment while maintaining the broad coverage capability of band offset.
Solution Approach 2:
The invention applies different offset values to different selected bands based on their specific characteristics. By selecting 4 consecutive bands that most benefit from offset adjustment and applying tailored offset values to each, the method achieves local optimization of pixel values in specific gray-scale regions, improving image quality where needed most.
2Reliability
If the band offset signals SAO parameters inclusive of both a determined band position and offset values corresponding to four respective bands for each block, then the decoder can reconstruct the image, but a large number of bits are transmitted as SAO parameters
Solution Approach 1:
The invention extracts only the essential SAO parameters needed for reconstruction. Instead of transmitting all possible band position information and offset values for every band, it transmits only the parameters corresponding to the 4 selected consecutive bands that require offset adjustment, reducing the parameter set to the minimum necessary for accurate reconstruction.
Solution Approach 2:
The invention transmits SAO parameters for only 4 selected bands rather than all 32 bands. This partial action approach transmits sufficient information to achieve the necessary image reconstruction quality while significantly reducing the total number of bits required, as most bands do not require offset adjustment.
3Ease of operation
If the currently utilized band offset with wide bandwidth is used, then the processing is simplified, but it cannot add or subtract offset values to proper targets when four consecutive bands other than the median value are selected, causing degradation in subjective image quality
Solution Approach 1:
The invention introduces dynamic selection of 4 consecutive bands based on the actual image content and characteristics. Rather than fixed processing of all bands or only median bands, the method dynamically identifies which 4 consecutive bands most benefit from offset adjustment, adapting to different image regions and content types to optimize image quality.
Solution Approach 2:
The invention changes the parameters of which bands receive offset adjustment based on image characteristics. By selecting 4 consecutive bands that most need adjustment rather than applying uniform processing, the method adapts the offset processing parameters to match the specific requirements of different image regions, improving subjective image quality.
Data Source
AI summary
An encoding method includes a counting process of dividing a gray-scale range of pixels of a coding unit block into a plurality of classes each having a predetermined number of gray-scale levels, and counting a number of pixels belonging to each of the plurality of classes, a class addition process of selecting top n classes (n: an integer greater than or equal to 1) having a largest number of counted pixels, and adding offset values to pixel values of pixels belonging to the n selected classes; and a first signaling process of signaling SAO parameters including the added offset values.